November 19, 2004
A study lead-authored by Michael Hout, sociology professor at UC Berkeley and chair of the university's graduate Sociology and Demography group, concludes that President Bush received a suspiciously high number of votes in Florida counties that used electronic voting. They estimate the "excess" number of Bush votes at 130,733 and according to the story in Wired News, they argue the odds that this number of excess votes was to due to chance alone are less than 1 in 1000.
Allow me to point out one potentially serious omission in this study.
Two words: Jewish voters.
Just three counties account for more than 100% of the excess. According to the Hout, et. al. statistical model, as recounted in the Wired News story, Palm Beach, Broward, and Miami-Dade counties produced approximately 150,000 excess Bush votes.
But the three counties in question have exceptionally high percentages of Jewish residents. According to this source (which, admittedly, uses six-year-old data, but it's hard to believe that there would be any big change with more current data), the West Palm Beach metropolitan area has the largest Jewish percentage of residents--16%--of any metropolitan area in the nation. Miami--Ft. Lauderdale, which spans Miami-Dade and Broward counties, is third-highest with 10.4%.
Do you think at least some Jewish voters who normally voted Democratic might have voted for President Bush? I do. (This story in the Ft. Lauderdale Sun-Sentinel is mildly supportive.) Do you think that this is a more reasonable (partial) explanation for the excess in these three counties, whereas the twelve other counties that used electronic voting collectively produced no "excess" Bush votes? I do. (I don't think that the number of Jewish voters is high enough to account for the entire 130,733 excess, but it could reasonably account for part of it.)
But the statistical models reported by the UC Berkeley authors on pages 6-10 of their study take no account of the proportion of Jewish voters and they include no variables that seem likely even to be significantly correlated with that characteristic.
To the authors' credit they have posted all their data on the Net. To check my hypothesis, all that would be required would be to enter a variable into their statistical model that estimated the Jewish voters in each Florida county. But I don't have that data at my fingertips and I have other pressing obligations, so I'll leave that as an "exercise for the reader".
Another possibly important omission is age. These three counties have a lot of retirees, probably more than the rest of Florida on average. But the authors' model does not include age, either.
(Thanks to Wizbang for blogging the story.)